Using deep learning for acoustic event classification: The case of natural disasters

Area of Science:

• Artificial Intelligence
• Machine Learning
• Signal Processing

Background:

• Natural disaster classification traditionally relies on visual or seismic data.
• Acoustic signals offer a complementary data source for disaster monitoring.
• Developing automated systems for disaster classification is crucial for timely response.

Purpose of the Study:

• To propose and evaluate deep learning models for natural disaster sound classification.
• To assess the effectiveness of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) classifiers.
• To demonstrate the potential of acoustic signals in machine learning-based disaster classification.

Main Methods:

• Utilized a dataset of 12,937 sound segments, truncated to 0.1 seconds.
• Trained two individual deep learning classifiers: a CNN and an LSTM.
• Employed machine learning techniques for analyzing acoustic signals.

Main Results:

• The CNN model achieved a classification accuracy of 99.96%.
• The LSTM model achieved a classification accuracy of 99.90%.
• Reported low misclassification rates (0.4% for CNN, 0.1% for LSTM), outperforming existing studies.

Conclusions:

• Acoustic signals are highly effective for natural disaster classification using machine learning.
• CNN and LSTM models provide accurate and reliable disaster classification.
• These classifiers represent a promising alternative for disaster classification and potential early detection.